1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 */ 24 25 /* 26 * This file contains the top half of the zfs directory structure 27 * implementation. The bottom half is in zap_leaf.c. 28 * 29 * The zdir is an extendable hash data structure. There is a table of 30 * pointers to buckets (zap_t->zd_data->zd_leafs). The buckets are 31 * each a constant size and hold a variable number of directory entries. 32 * The buckets (aka "leaf nodes") are implemented in zap_leaf.c. 33 * 34 * The pointer table holds a power of 2 number of pointers. 35 * (1<<zap_t->zd_data->zd_phys->zd_prefix_len). The bucket pointed to 36 * by the pointer at index i in the table holds entries whose hash value 37 * has a zd_prefix_len - bit prefix 38 */ 39 40 #include <sys/spa.h> 41 #include <sys/dmu.h> 42 #include <sys/zfs_context.h> 43 #include <sys/zfs_znode.h> 44 #include <sys/fs/zfs.h> 45 #include <sys/zap.h> 46 #include <sys/refcount.h> 47 #include <sys/zap_impl.h> 48 #include <sys/zap_leaf.h> 49 50 int fzap_default_block_shift = 14; /* 16k blocksize */ 51 52 static void zap_leaf_pageout(dmu_buf_t *db, void *vl); 53 static uint64_t zap_allocate_blocks(zap_t *zap, int nblocks); 54 55 56 void 57 fzap_byteswap(void *vbuf, size_t size) 58 { 59 uint64_t block_type; 60 61 block_type = *(uint64_t *)vbuf; 62 63 if (block_type == ZBT_LEAF || block_type == BSWAP_64(ZBT_LEAF)) 64 zap_leaf_byteswap(vbuf, size); 65 else { 66 /* it's a ptrtbl block */ 67 byteswap_uint64_array(vbuf, size); 68 } 69 } 70 71 void 72 fzap_upgrade(zap_t *zap, dmu_tx_t *tx, zap_flags_t flags) 73 { 74 dmu_buf_t *db; 75 zap_leaf_t *l; 76 int i; 77 zap_phys_t *zp; 78 79 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 80 zap->zap_ismicro = FALSE; 81 82 (void) dmu_buf_update_user(zap->zap_dbuf, zap, zap, 83 &zap->zap_f.zap_phys, zap_evict); 84 85 mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0); 86 zap->zap_f.zap_block_shift = highbit(zap->zap_dbuf->db_size) - 1; 87 88 zp = zap->zap_f.zap_phys; 89 /* 90 * explicitly zero it since it might be coming from an 91 * initialized microzap 92 */ 93 bzero(zap->zap_dbuf->db_data, zap->zap_dbuf->db_size); 94 zp->zap_block_type = ZBT_HEADER; 95 zp->zap_magic = ZAP_MAGIC; 96 97 zp->zap_ptrtbl.zt_shift = ZAP_EMBEDDED_PTRTBL_SHIFT(zap); 98 99 zp->zap_freeblk = 2; /* block 1 will be the first leaf */ 100 zp->zap_num_leafs = 1; 101 zp->zap_num_entries = 0; 102 zp->zap_salt = zap->zap_salt; 103 zp->zap_normflags = zap->zap_normflags; 104 zp->zap_flags = flags; 105 106 /* block 1 will be the first leaf */ 107 for (i = 0; i < (1<<zp->zap_ptrtbl.zt_shift); i++) 108 ZAP_EMBEDDED_PTRTBL_ENT(zap, i) = 1; 109 110 /* 111 * set up block 1 - the first leaf 112 */ 113 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object, 114 1<<FZAP_BLOCK_SHIFT(zap), FTAG, &db, DMU_READ_NO_PREFETCH)); 115 dmu_buf_will_dirty(db, tx); 116 117 l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP); 118 l->l_dbuf = db; 119 l->l_phys = db->db_data; 120 121 zap_leaf_init(l, zp->zap_normflags != 0); 122 123 kmem_free(l, sizeof (zap_leaf_t)); 124 dmu_buf_rele(db, FTAG); 125 } 126 127 static int 128 zap_tryupgradedir(zap_t *zap, dmu_tx_t *tx) 129 { 130 if (RW_WRITE_HELD(&zap->zap_rwlock)) 131 return (1); 132 if (rw_tryupgrade(&zap->zap_rwlock)) { 133 dmu_buf_will_dirty(zap->zap_dbuf, tx); 134 return (1); 135 } 136 return (0); 137 } 138 139 /* 140 * Generic routines for dealing with the pointer & cookie tables. 141 */ 142 143 static int 144 zap_table_grow(zap_t *zap, zap_table_phys_t *tbl, 145 void (*transfer_func)(const uint64_t *src, uint64_t *dst, int n), 146 dmu_tx_t *tx) 147 { 148 uint64_t b, newblk; 149 dmu_buf_t *db_old, *db_new; 150 int err; 151 int bs = FZAP_BLOCK_SHIFT(zap); 152 int hepb = 1<<(bs-4); 153 /* hepb = half the number of entries in a block */ 154 155 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 156 ASSERT(tbl->zt_blk != 0); 157 ASSERT(tbl->zt_numblks > 0); 158 159 if (tbl->zt_nextblk != 0) { 160 newblk = tbl->zt_nextblk; 161 } else { 162 newblk = zap_allocate_blocks(zap, tbl->zt_numblks * 2); 163 tbl->zt_nextblk = newblk; 164 ASSERT3U(tbl->zt_blks_copied, ==, 0); 165 dmu_prefetch(zap->zap_objset, zap->zap_object, 166 tbl->zt_blk << bs, tbl->zt_numblks << bs); 167 } 168 169 /* 170 * Copy the ptrtbl from the old to new location. 171 */ 172 173 b = tbl->zt_blks_copied; 174 err = dmu_buf_hold(zap->zap_objset, zap->zap_object, 175 (tbl->zt_blk + b) << bs, FTAG, &db_old, DMU_READ_NO_PREFETCH); 176 if (err) 177 return (err); 178 179 /* first half of entries in old[b] go to new[2*b+0] */ 180 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object, 181 (newblk + 2*b+0) << bs, FTAG, &db_new, DMU_READ_NO_PREFETCH)); 182 dmu_buf_will_dirty(db_new, tx); 183 transfer_func(db_old->db_data, db_new->db_data, hepb); 184 dmu_buf_rele(db_new, FTAG); 185 186 /* second half of entries in old[b] go to new[2*b+1] */ 187 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object, 188 (newblk + 2*b+1) << bs, FTAG, &db_new, DMU_READ_NO_PREFETCH)); 189 dmu_buf_will_dirty(db_new, tx); 190 transfer_func((uint64_t *)db_old->db_data + hepb, 191 db_new->db_data, hepb); 192 dmu_buf_rele(db_new, FTAG); 193 194 dmu_buf_rele(db_old, FTAG); 195 196 tbl->zt_blks_copied++; 197 198 dprintf("copied block %llu of %llu\n", 199 tbl->zt_blks_copied, tbl->zt_numblks); 200 201 if (tbl->zt_blks_copied == tbl->zt_numblks) { 202 (void) dmu_free_range(zap->zap_objset, zap->zap_object, 203 tbl->zt_blk << bs, tbl->zt_numblks << bs, tx); 204 205 tbl->zt_blk = newblk; 206 tbl->zt_numblks *= 2; 207 tbl->zt_shift++; 208 tbl->zt_nextblk = 0; 209 tbl->zt_blks_copied = 0; 210 211 dprintf("finished; numblocks now %llu (%lluk entries)\n", 212 tbl->zt_numblks, 1<<(tbl->zt_shift-10)); 213 } 214 215 return (0); 216 } 217 218 static int 219 zap_table_store(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t val, 220 dmu_tx_t *tx) 221 { 222 int err; 223 uint64_t blk, off; 224 int bs = FZAP_BLOCK_SHIFT(zap); 225 dmu_buf_t *db; 226 227 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 228 ASSERT(tbl->zt_blk != 0); 229 230 dprintf("storing %llx at index %llx\n", val, idx); 231 232 blk = idx >> (bs-3); 233 off = idx & ((1<<(bs-3))-1); 234 235 err = dmu_buf_hold(zap->zap_objset, zap->zap_object, 236 (tbl->zt_blk + blk) << bs, FTAG, &db, DMU_READ_NO_PREFETCH); 237 if (err) 238 return (err); 239 dmu_buf_will_dirty(db, tx); 240 241 if (tbl->zt_nextblk != 0) { 242 uint64_t idx2 = idx * 2; 243 uint64_t blk2 = idx2 >> (bs-3); 244 uint64_t off2 = idx2 & ((1<<(bs-3))-1); 245 dmu_buf_t *db2; 246 247 err = dmu_buf_hold(zap->zap_objset, zap->zap_object, 248 (tbl->zt_nextblk + blk2) << bs, FTAG, &db2, 249 DMU_READ_NO_PREFETCH); 250 if (err) { 251 dmu_buf_rele(db, FTAG); 252 return (err); 253 } 254 dmu_buf_will_dirty(db2, tx); 255 ((uint64_t *)db2->db_data)[off2] = val; 256 ((uint64_t *)db2->db_data)[off2+1] = val; 257 dmu_buf_rele(db2, FTAG); 258 } 259 260 ((uint64_t *)db->db_data)[off] = val; 261 dmu_buf_rele(db, FTAG); 262 263 return (0); 264 } 265 266 static int 267 zap_table_load(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t *valp) 268 { 269 uint64_t blk, off; 270 int err; 271 dmu_buf_t *db; 272 int bs = FZAP_BLOCK_SHIFT(zap); 273 274 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 275 276 blk = idx >> (bs-3); 277 off = idx & ((1<<(bs-3))-1); 278 279 err = dmu_buf_hold(zap->zap_objset, zap->zap_object, 280 (tbl->zt_blk + blk) << bs, FTAG, &db, DMU_READ_NO_PREFETCH); 281 if (err) 282 return (err); 283 *valp = ((uint64_t *)db->db_data)[off]; 284 dmu_buf_rele(db, FTAG); 285 286 if (tbl->zt_nextblk != 0) { 287 /* 288 * read the nextblk for the sake of i/o error checking, 289 * so that zap_table_load() will catch errors for 290 * zap_table_store. 291 */ 292 blk = (idx*2) >> (bs-3); 293 294 err = dmu_buf_hold(zap->zap_objset, zap->zap_object, 295 (tbl->zt_nextblk + blk) << bs, FTAG, &db, 296 DMU_READ_NO_PREFETCH); 297 dmu_buf_rele(db, FTAG); 298 } 299 return (err); 300 } 301 302 /* 303 * Routines for growing the ptrtbl. 304 */ 305 306 static void 307 zap_ptrtbl_transfer(const uint64_t *src, uint64_t *dst, int n) 308 { 309 int i; 310 for (i = 0; i < n; i++) { 311 uint64_t lb = src[i]; 312 dst[2*i+0] = lb; 313 dst[2*i+1] = lb; 314 } 315 } 316 317 static int 318 zap_grow_ptrtbl(zap_t *zap, dmu_tx_t *tx) 319 { 320 /* 321 * The pointer table should never use more hash bits than we 322 * have (otherwise we'd be using useless zero bits to index it). 323 * If we are within 2 bits of running out, stop growing, since 324 * this is already an aberrant condition. 325 */ 326 if (zap->zap_f.zap_phys->zap_ptrtbl.zt_shift >= zap_hashbits(zap) - 2) 327 return (ENOSPC); 328 329 if (zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks == 0) { 330 /* 331 * We are outgrowing the "embedded" ptrtbl (the one 332 * stored in the header block). Give it its own entire 333 * block, which will double the size of the ptrtbl. 334 */ 335 uint64_t newblk; 336 dmu_buf_t *db_new; 337 int err; 338 339 ASSERT3U(zap->zap_f.zap_phys->zap_ptrtbl.zt_shift, ==, 340 ZAP_EMBEDDED_PTRTBL_SHIFT(zap)); 341 ASSERT3U(zap->zap_f.zap_phys->zap_ptrtbl.zt_blk, ==, 0); 342 343 newblk = zap_allocate_blocks(zap, 1); 344 err = dmu_buf_hold(zap->zap_objset, zap->zap_object, 345 newblk << FZAP_BLOCK_SHIFT(zap), FTAG, &db_new, 346 DMU_READ_NO_PREFETCH); 347 if (err) 348 return (err); 349 dmu_buf_will_dirty(db_new, tx); 350 zap_ptrtbl_transfer(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), 351 db_new->db_data, 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap)); 352 dmu_buf_rele(db_new, FTAG); 353 354 zap->zap_f.zap_phys->zap_ptrtbl.zt_blk = newblk; 355 zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks = 1; 356 zap->zap_f.zap_phys->zap_ptrtbl.zt_shift++; 357 358 ASSERT3U(1ULL << zap->zap_f.zap_phys->zap_ptrtbl.zt_shift, ==, 359 zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks << 360 (FZAP_BLOCK_SHIFT(zap)-3)); 361 362 return (0); 363 } else { 364 return (zap_table_grow(zap, &zap->zap_f.zap_phys->zap_ptrtbl, 365 zap_ptrtbl_transfer, tx)); 366 } 367 } 368 369 static void 370 zap_increment_num_entries(zap_t *zap, int delta, dmu_tx_t *tx) 371 { 372 dmu_buf_will_dirty(zap->zap_dbuf, tx); 373 mutex_enter(&zap->zap_f.zap_num_entries_mtx); 374 ASSERT(delta > 0 || zap->zap_f.zap_phys->zap_num_entries >= -delta); 375 zap->zap_f.zap_phys->zap_num_entries += delta; 376 mutex_exit(&zap->zap_f.zap_num_entries_mtx); 377 } 378 379 static uint64_t 380 zap_allocate_blocks(zap_t *zap, int nblocks) 381 { 382 uint64_t newblk; 383 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 384 newblk = zap->zap_f.zap_phys->zap_freeblk; 385 zap->zap_f.zap_phys->zap_freeblk += nblocks; 386 return (newblk); 387 } 388 389 static zap_leaf_t * 390 zap_create_leaf(zap_t *zap, dmu_tx_t *tx) 391 { 392 void *winner; 393 zap_leaf_t *l = kmem_alloc(sizeof (zap_leaf_t), KM_SLEEP); 394 395 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 396 397 rw_init(&l->l_rwlock, 0, 0, 0); 398 rw_enter(&l->l_rwlock, RW_WRITER); 399 l->l_blkid = zap_allocate_blocks(zap, 1); 400 l->l_dbuf = NULL; 401 l->l_phys = NULL; 402 403 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object, 404 l->l_blkid << FZAP_BLOCK_SHIFT(zap), NULL, &l->l_dbuf, 405 DMU_READ_NO_PREFETCH)); 406 winner = dmu_buf_set_user(l->l_dbuf, l, &l->l_phys, zap_leaf_pageout); 407 ASSERT(winner == NULL); 408 dmu_buf_will_dirty(l->l_dbuf, tx); 409 410 zap_leaf_init(l, zap->zap_normflags != 0); 411 412 zap->zap_f.zap_phys->zap_num_leafs++; 413 414 return (l); 415 } 416 417 int 418 fzap_count(zap_t *zap, uint64_t *count) 419 { 420 ASSERT(!zap->zap_ismicro); 421 mutex_enter(&zap->zap_f.zap_num_entries_mtx); /* unnecessary */ 422 *count = zap->zap_f.zap_phys->zap_num_entries; 423 mutex_exit(&zap->zap_f.zap_num_entries_mtx); 424 return (0); 425 } 426 427 /* 428 * Routines for obtaining zap_leaf_t's 429 */ 430 431 void 432 zap_put_leaf(zap_leaf_t *l) 433 { 434 rw_exit(&l->l_rwlock); 435 dmu_buf_rele(l->l_dbuf, NULL); 436 } 437 438 _NOTE(ARGSUSED(0)) 439 static void 440 zap_leaf_pageout(dmu_buf_t *db, void *vl) 441 { 442 zap_leaf_t *l = vl; 443 444 rw_destroy(&l->l_rwlock); 445 kmem_free(l, sizeof (zap_leaf_t)); 446 } 447 448 static zap_leaf_t * 449 zap_open_leaf(uint64_t blkid, dmu_buf_t *db) 450 { 451 zap_leaf_t *l, *winner; 452 453 ASSERT(blkid != 0); 454 455 l = kmem_alloc(sizeof (zap_leaf_t), KM_SLEEP); 456 rw_init(&l->l_rwlock, 0, 0, 0); 457 rw_enter(&l->l_rwlock, RW_WRITER); 458 l->l_blkid = blkid; 459 l->l_bs = highbit(db->db_size)-1; 460 l->l_dbuf = db; 461 l->l_phys = NULL; 462 463 winner = dmu_buf_set_user(db, l, &l->l_phys, zap_leaf_pageout); 464 465 rw_exit(&l->l_rwlock); 466 if (winner != NULL) { 467 /* someone else set it first */ 468 zap_leaf_pageout(NULL, l); 469 l = winner; 470 } 471 472 /* 473 * lhr_pad was previously used for the next leaf in the leaf 474 * chain. There should be no chained leafs (as we have removed 475 * support for them). 476 */ 477 ASSERT3U(l->l_phys->l_hdr.lh_pad1, ==, 0); 478 479 /* 480 * There should be more hash entries than there can be 481 * chunks to put in the hash table 482 */ 483 ASSERT3U(ZAP_LEAF_HASH_NUMENTRIES(l), >, ZAP_LEAF_NUMCHUNKS(l) / 3); 484 485 /* The chunks should begin at the end of the hash table */ 486 ASSERT3P(&ZAP_LEAF_CHUNK(l, 0), ==, 487 &l->l_phys->l_hash[ZAP_LEAF_HASH_NUMENTRIES(l)]); 488 489 /* The chunks should end at the end of the block */ 490 ASSERT3U((uintptr_t)&ZAP_LEAF_CHUNK(l, ZAP_LEAF_NUMCHUNKS(l)) - 491 (uintptr_t)l->l_phys, ==, l->l_dbuf->db_size); 492 493 return (l); 494 } 495 496 static int 497 zap_get_leaf_byblk(zap_t *zap, uint64_t blkid, dmu_tx_t *tx, krw_t lt, 498 zap_leaf_t **lp) 499 { 500 dmu_buf_t *db; 501 zap_leaf_t *l; 502 int bs = FZAP_BLOCK_SHIFT(zap); 503 int err; 504 505 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 506 507 err = dmu_buf_hold(zap->zap_objset, zap->zap_object, 508 blkid << bs, NULL, &db, DMU_READ_NO_PREFETCH); 509 if (err) 510 return (err); 511 512 ASSERT3U(db->db_object, ==, zap->zap_object); 513 ASSERT3U(db->db_offset, ==, blkid << bs); 514 ASSERT3U(db->db_size, ==, 1 << bs); 515 ASSERT(blkid != 0); 516 517 l = dmu_buf_get_user(db); 518 519 if (l == NULL) 520 l = zap_open_leaf(blkid, db); 521 522 rw_enter(&l->l_rwlock, lt); 523 /* 524 * Must lock before dirtying, otherwise l->l_phys could change, 525 * causing ASSERT below to fail. 526 */ 527 if (lt == RW_WRITER) 528 dmu_buf_will_dirty(db, tx); 529 ASSERT3U(l->l_blkid, ==, blkid); 530 ASSERT3P(l->l_dbuf, ==, db); 531 ASSERT3P(l->l_phys, ==, l->l_dbuf->db_data); 532 ASSERT3U(l->l_phys->l_hdr.lh_block_type, ==, ZBT_LEAF); 533 ASSERT3U(l->l_phys->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC); 534 535 *lp = l; 536 return (0); 537 } 538 539 static int 540 zap_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t *valp) 541 { 542 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 543 544 if (zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks == 0) { 545 ASSERT3U(idx, <, 546 (1ULL << zap->zap_f.zap_phys->zap_ptrtbl.zt_shift)); 547 *valp = ZAP_EMBEDDED_PTRTBL_ENT(zap, idx); 548 return (0); 549 } else { 550 return (zap_table_load(zap, &zap->zap_f.zap_phys->zap_ptrtbl, 551 idx, valp)); 552 } 553 } 554 555 static int 556 zap_set_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t blk, dmu_tx_t *tx) 557 { 558 ASSERT(tx != NULL); 559 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 560 561 if (zap->zap_f.zap_phys->zap_ptrtbl.zt_blk == 0) { 562 ZAP_EMBEDDED_PTRTBL_ENT(zap, idx) = blk; 563 return (0); 564 } else { 565 return (zap_table_store(zap, &zap->zap_f.zap_phys->zap_ptrtbl, 566 idx, blk, tx)); 567 } 568 } 569 570 static int 571 zap_deref_leaf(zap_t *zap, uint64_t h, dmu_tx_t *tx, krw_t lt, zap_leaf_t **lp) 572 { 573 uint64_t idx, blk; 574 int err; 575 576 ASSERT(zap->zap_dbuf == NULL || 577 zap->zap_f.zap_phys == zap->zap_dbuf->db_data); 578 ASSERT3U(zap->zap_f.zap_phys->zap_magic, ==, ZAP_MAGIC); 579 idx = ZAP_HASH_IDX(h, zap->zap_f.zap_phys->zap_ptrtbl.zt_shift); 580 err = zap_idx_to_blk(zap, idx, &blk); 581 if (err != 0) 582 return (err); 583 err = zap_get_leaf_byblk(zap, blk, tx, lt, lp); 584 585 ASSERT(err || ZAP_HASH_IDX(h, (*lp)->l_phys->l_hdr.lh_prefix_len) == 586 (*lp)->l_phys->l_hdr.lh_prefix); 587 return (err); 588 } 589 590 static int 591 zap_expand_leaf(zap_name_t *zn, zap_leaf_t *l, dmu_tx_t *tx, zap_leaf_t **lp) 592 { 593 zap_t *zap = zn->zn_zap; 594 uint64_t hash = zn->zn_hash; 595 zap_leaf_t *nl; 596 int prefix_diff, i, err; 597 uint64_t sibling; 598 int old_prefix_len = l->l_phys->l_hdr.lh_prefix_len; 599 600 ASSERT3U(old_prefix_len, <=, zap->zap_f.zap_phys->zap_ptrtbl.zt_shift); 601 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 602 603 ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==, 604 l->l_phys->l_hdr.lh_prefix); 605 606 if (zap_tryupgradedir(zap, tx) == 0 || 607 old_prefix_len == zap->zap_f.zap_phys->zap_ptrtbl.zt_shift) { 608 /* We failed to upgrade, or need to grow the pointer table */ 609 objset_t *os = zap->zap_objset; 610 uint64_t object = zap->zap_object; 611 612 zap_put_leaf(l); 613 zap_unlockdir(zap); 614 err = zap_lockdir(os, object, tx, RW_WRITER, 615 FALSE, FALSE, &zn->zn_zap); 616 zap = zn->zn_zap; 617 if (err) 618 return (err); 619 ASSERT(!zap->zap_ismicro); 620 621 while (old_prefix_len == 622 zap->zap_f.zap_phys->zap_ptrtbl.zt_shift) { 623 err = zap_grow_ptrtbl(zap, tx); 624 if (err) 625 return (err); 626 } 627 628 err = zap_deref_leaf(zap, hash, tx, RW_WRITER, &l); 629 if (err) 630 return (err); 631 632 if (l->l_phys->l_hdr.lh_prefix_len != old_prefix_len) { 633 /* it split while our locks were down */ 634 *lp = l; 635 return (0); 636 } 637 } 638 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock)); 639 ASSERT3U(old_prefix_len, <, zap->zap_f.zap_phys->zap_ptrtbl.zt_shift); 640 ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==, 641 l->l_phys->l_hdr.lh_prefix); 642 643 prefix_diff = zap->zap_f.zap_phys->zap_ptrtbl.zt_shift - 644 (old_prefix_len + 1); 645 sibling = (ZAP_HASH_IDX(hash, old_prefix_len + 1) | 1) << prefix_diff; 646 647 /* check for i/o errors before doing zap_leaf_split */ 648 for (i = 0; i < (1ULL<<prefix_diff); i++) { 649 uint64_t blk; 650 err = zap_idx_to_blk(zap, sibling+i, &blk); 651 if (err) 652 return (err); 653 ASSERT3U(blk, ==, l->l_blkid); 654 } 655 656 nl = zap_create_leaf(zap, tx); 657 zap_leaf_split(l, nl, zap->zap_normflags != 0); 658 659 /* set sibling pointers */ 660 for (i = 0; i < (1ULL<<prefix_diff); i++) { 661 err = zap_set_idx_to_blk(zap, sibling+i, nl->l_blkid, tx); 662 ASSERT3U(err, ==, 0); /* we checked for i/o errors above */ 663 } 664 665 if (hash & (1ULL << (64 - l->l_phys->l_hdr.lh_prefix_len))) { 666 /* we want the sibling */ 667 zap_put_leaf(l); 668 *lp = nl; 669 } else { 670 zap_put_leaf(nl); 671 *lp = l; 672 } 673 674 return (0); 675 } 676 677 static void 678 zap_put_leaf_maybe_grow_ptrtbl(zap_name_t *zn, zap_leaf_t *l, dmu_tx_t *tx) 679 { 680 zap_t *zap = zn->zn_zap; 681 int shift = zap->zap_f.zap_phys->zap_ptrtbl.zt_shift; 682 int leaffull = (l->l_phys->l_hdr.lh_prefix_len == shift && 683 l->l_phys->l_hdr.lh_nfree < ZAP_LEAF_LOW_WATER); 684 685 zap_put_leaf(l); 686 687 if (leaffull || zap->zap_f.zap_phys->zap_ptrtbl.zt_nextblk) { 688 int err; 689 690 /* 691 * We are in the middle of growing the pointer table, or 692 * this leaf will soon make us grow it. 693 */ 694 if (zap_tryupgradedir(zap, tx) == 0) { 695 objset_t *os = zap->zap_objset; 696 uint64_t zapobj = zap->zap_object; 697 698 zap_unlockdir(zap); 699 err = zap_lockdir(os, zapobj, tx, 700 RW_WRITER, FALSE, FALSE, &zn->zn_zap); 701 zap = zn->zn_zap; 702 if (err) 703 return; 704 } 705 706 /* could have finished growing while our locks were down */ 707 if (zap->zap_f.zap_phys->zap_ptrtbl.zt_shift == shift) 708 (void) zap_grow_ptrtbl(zap, tx); 709 } 710 } 711 712 static int 713 fzap_checkname(zap_name_t *zn) 714 { 715 if (zn->zn_key_orig_numints * zn->zn_key_intlen > ZAP_MAXNAMELEN) 716 return (ENAMETOOLONG); 717 return (0); 718 } 719 720 static int 721 fzap_checksize(uint64_t integer_size, uint64_t num_integers) 722 { 723 /* Only integer sizes supported by C */ 724 switch (integer_size) { 725 case 1: 726 case 2: 727 case 4: 728 case 8: 729 break; 730 default: 731 return (EINVAL); 732 } 733 734 if (integer_size * num_integers > ZAP_MAXVALUELEN) 735 return (E2BIG); 736 737 return (0); 738 } 739 740 static int 741 fzap_check(zap_name_t *zn, uint64_t integer_size, uint64_t num_integers) 742 { 743 int err; 744 745 if ((err = fzap_checkname(zn)) != 0) 746 return (err); 747 return (fzap_checksize(integer_size, num_integers)); 748 } 749 750 /* 751 * Routines for manipulating attributes. 752 */ 753 int 754 fzap_lookup(zap_name_t *zn, 755 uint64_t integer_size, uint64_t num_integers, void *buf, 756 char *realname, int rn_len, boolean_t *ncp) 757 { 758 zap_leaf_t *l; 759 int err; 760 zap_entry_handle_t zeh; 761 762 if ((err = fzap_checkname(zn)) != 0) 763 return (err); 764 765 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l); 766 if (err != 0) 767 return (err); 768 err = zap_leaf_lookup(l, zn, &zeh); 769 if (err == 0) { 770 if ((err = fzap_checksize(integer_size, num_integers)) != 0) { 771 zap_put_leaf(l); 772 return (err); 773 } 774 775 err = zap_entry_read(&zeh, integer_size, num_integers, buf); 776 (void) zap_entry_read_name(zn->zn_zap, &zeh, rn_len, realname); 777 if (ncp) { 778 *ncp = zap_entry_normalization_conflict(&zeh, 779 zn, NULL, zn->zn_zap); 780 } 781 } 782 783 zap_put_leaf(l); 784 return (err); 785 } 786 787 int 788 fzap_add_cd(zap_name_t *zn, 789 uint64_t integer_size, uint64_t num_integers, 790 const void *val, uint32_t cd, dmu_tx_t *tx) 791 { 792 zap_leaf_t *l; 793 int err; 794 zap_entry_handle_t zeh; 795 zap_t *zap = zn->zn_zap; 796 797 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 798 ASSERT(!zap->zap_ismicro); 799 ASSERT(fzap_check(zn, integer_size, num_integers) == 0); 800 801 err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l); 802 if (err != 0) 803 return (err); 804 retry: 805 err = zap_leaf_lookup(l, zn, &zeh); 806 if (err == 0) { 807 err = EEXIST; 808 goto out; 809 } 810 if (err != ENOENT) 811 goto out; 812 813 err = zap_entry_create(l, zn, cd, 814 integer_size, num_integers, val, &zeh); 815 816 if (err == 0) { 817 zap_increment_num_entries(zap, 1, tx); 818 } else if (err == EAGAIN) { 819 err = zap_expand_leaf(zn, l, tx, &l); 820 zap = zn->zn_zap; /* zap_expand_leaf() may change zap */ 821 if (err == 0) 822 goto retry; 823 } 824 825 out: 826 if (zap != NULL) 827 zap_put_leaf_maybe_grow_ptrtbl(zn, l, tx); 828 return (err); 829 } 830 831 int 832 fzap_add(zap_name_t *zn, 833 uint64_t integer_size, uint64_t num_integers, 834 const void *val, dmu_tx_t *tx) 835 { 836 int err = fzap_check(zn, integer_size, num_integers); 837 if (err != 0) 838 return (err); 839 840 return (fzap_add_cd(zn, integer_size, num_integers, 841 val, ZAP_NEED_CD, tx)); 842 } 843 844 int 845 fzap_update(zap_name_t *zn, 846 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx) 847 { 848 zap_leaf_t *l; 849 int err, create; 850 zap_entry_handle_t zeh; 851 zap_t *zap = zn->zn_zap; 852 853 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock)); 854 err = fzap_check(zn, integer_size, num_integers); 855 if (err != 0) 856 return (err); 857 858 err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l); 859 if (err != 0) 860 return (err); 861 retry: 862 err = zap_leaf_lookup(l, zn, &zeh); 863 create = (err == ENOENT); 864 ASSERT(err == 0 || err == ENOENT); 865 866 if (create) { 867 err = zap_entry_create(l, zn, ZAP_NEED_CD, 868 integer_size, num_integers, val, &zeh); 869 if (err == 0) 870 zap_increment_num_entries(zap, 1, tx); 871 } else { 872 err = zap_entry_update(&zeh, integer_size, num_integers, val); 873 } 874 875 if (err == EAGAIN) { 876 err = zap_expand_leaf(zn, l, tx, &l); 877 zap = zn->zn_zap; /* zap_expand_leaf() may change zap */ 878 if (err == 0) 879 goto retry; 880 } 881 882 if (zap != NULL) 883 zap_put_leaf_maybe_grow_ptrtbl(zn, l, tx); 884 return (err); 885 } 886 887 int 888 fzap_length(zap_name_t *zn, 889 uint64_t *integer_size, uint64_t *num_integers) 890 { 891 zap_leaf_t *l; 892 int err; 893 zap_entry_handle_t zeh; 894 895 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l); 896 if (err != 0) 897 return (err); 898 err = zap_leaf_lookup(l, zn, &zeh); 899 if (err != 0) 900 goto out; 901 902 if (integer_size) 903 *integer_size = zeh.zeh_integer_size; 904 if (num_integers) 905 *num_integers = zeh.zeh_num_integers; 906 out: 907 zap_put_leaf(l); 908 return (err); 909 } 910 911 int 912 fzap_remove(zap_name_t *zn, dmu_tx_t *tx) 913 { 914 zap_leaf_t *l; 915 int err; 916 zap_entry_handle_t zeh; 917 918 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, tx, RW_WRITER, &l); 919 if (err != 0) 920 return (err); 921 err = zap_leaf_lookup(l, zn, &zeh); 922 if (err == 0) { 923 zap_entry_remove(&zeh); 924 zap_increment_num_entries(zn->zn_zap, -1, tx); 925 } 926 zap_put_leaf(l); 927 return (err); 928 } 929 930 /* 931 * Helper functions for consumers. 932 */ 933 934 int 935 zap_value_search(objset_t *os, uint64_t zapobj, uint64_t value, uint64_t mask, 936 char *name) 937 { 938 zap_cursor_t zc; 939 zap_attribute_t *za; 940 int err; 941 942 if (mask == 0) 943 mask = -1ULL; 944 945 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 946 for (zap_cursor_init(&zc, os, zapobj); 947 (err = zap_cursor_retrieve(&zc, za)) == 0; 948 zap_cursor_advance(&zc)) { 949 if ((za->za_first_integer & mask) == (value & mask)) { 950 (void) strcpy(name, za->za_name); 951 break; 952 } 953 } 954 zap_cursor_fini(&zc); 955 kmem_free(za, sizeof (zap_attribute_t)); 956 return (err); 957 } 958 959 int 960 zap_join(objset_t *os, uint64_t fromobj, uint64_t intoobj, dmu_tx_t *tx) 961 { 962 zap_cursor_t zc; 963 zap_attribute_t za; 964 int err; 965 966 for (zap_cursor_init(&zc, os, fromobj); 967 zap_cursor_retrieve(&zc, &za) == 0; 968 (void) zap_cursor_advance(&zc)) { 969 if (za.za_integer_length != 8 || za.za_num_integers != 1) 970 return (EINVAL); 971 err = zap_add(os, intoobj, za.za_name, 972 8, 1, &za.za_first_integer, tx); 973 if (err) 974 return (err); 975 } 976 zap_cursor_fini(&zc); 977 return (0); 978 } 979 980 int 981 zap_add_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx) 982 { 983 char name[20]; 984 985 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value); 986 return (zap_add(os, obj, name, 8, 1, &value, tx)); 987 } 988 989 int 990 zap_remove_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx) 991 { 992 char name[20]; 993 994 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value); 995 return (zap_remove(os, obj, name, tx)); 996 } 997 998 int 999 zap_lookup_int(objset_t *os, uint64_t obj, uint64_t value) 1000 { 1001 char name[20]; 1002 1003 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value); 1004 return (zap_lookup(os, obj, name, 8, 1, &value)); 1005 } 1006 1007 int 1008 zap_increment_int(objset_t *os, uint64_t obj, uint64_t key, int64_t delta, 1009 dmu_tx_t *tx) 1010 { 1011 char name[20]; 1012 uint64_t value = 0; 1013 int err; 1014 1015 if (delta == 0) 1016 return (0); 1017 1018 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key); 1019 err = zap_lookup(os, obj, name, 8, 1, &value); 1020 if (err != 0 && err != ENOENT) 1021 return (err); 1022 value += delta; 1023 if (value == 0) 1024 err = zap_remove(os, obj, name, tx); 1025 else 1026 err = zap_update(os, obj, name, 8, 1, &value, tx); 1027 return (err); 1028 } 1029 1030 1031 /* 1032 * Routines for iterating over the attributes. 1033 */ 1034 1035 int 1036 fzap_cursor_retrieve(zap_t *zap, zap_cursor_t *zc, zap_attribute_t *za) 1037 { 1038 int err = ENOENT; 1039 zap_entry_handle_t zeh; 1040 zap_leaf_t *l; 1041 1042 /* retrieve the next entry at or after zc_hash/zc_cd */ 1043 /* if no entry, return ENOENT */ 1044 1045 if (zc->zc_leaf && 1046 (ZAP_HASH_IDX(zc->zc_hash, 1047 zc->zc_leaf->l_phys->l_hdr.lh_prefix_len) != 1048 zc->zc_leaf->l_phys->l_hdr.lh_prefix)) { 1049 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER); 1050 zap_put_leaf(zc->zc_leaf); 1051 zc->zc_leaf = NULL; 1052 } 1053 1054 again: 1055 if (zc->zc_leaf == NULL) { 1056 err = zap_deref_leaf(zap, zc->zc_hash, NULL, RW_READER, 1057 &zc->zc_leaf); 1058 if (err != 0) 1059 return (err); 1060 } else { 1061 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER); 1062 } 1063 l = zc->zc_leaf; 1064 1065 err = zap_leaf_lookup_closest(l, zc->zc_hash, zc->zc_cd, &zeh); 1066 1067 if (err == ENOENT) { 1068 uint64_t nocare = 1069 (1ULL << (64 - l->l_phys->l_hdr.lh_prefix_len)) - 1; 1070 zc->zc_hash = (zc->zc_hash & ~nocare) + nocare + 1; 1071 zc->zc_cd = 0; 1072 if (l->l_phys->l_hdr.lh_prefix_len == 0 || zc->zc_hash == 0) { 1073 zc->zc_hash = -1ULL; 1074 } else { 1075 zap_put_leaf(zc->zc_leaf); 1076 zc->zc_leaf = NULL; 1077 goto again; 1078 } 1079 } 1080 1081 if (err == 0) { 1082 zc->zc_hash = zeh.zeh_hash; 1083 zc->zc_cd = zeh.zeh_cd; 1084 za->za_integer_length = zeh.zeh_integer_size; 1085 za->za_num_integers = zeh.zeh_num_integers; 1086 if (zeh.zeh_num_integers == 0) { 1087 za->za_first_integer = 0; 1088 } else { 1089 err = zap_entry_read(&zeh, 8, 1, &za->za_first_integer); 1090 ASSERT(err == 0 || err == EOVERFLOW); 1091 } 1092 err = zap_entry_read_name(zap, &zeh, 1093 sizeof (za->za_name), za->za_name); 1094 ASSERT(err == 0); 1095 1096 za->za_normalization_conflict = 1097 zap_entry_normalization_conflict(&zeh, 1098 NULL, za->za_name, zap); 1099 } 1100 rw_exit(&zc->zc_leaf->l_rwlock); 1101 return (err); 1102 } 1103 1104 1105 static void 1106 zap_stats_ptrtbl(zap_t *zap, uint64_t *tbl, int len, zap_stats_t *zs) 1107 { 1108 int i, err; 1109 uint64_t lastblk = 0; 1110 1111 /* 1112 * NB: if a leaf has more pointers than an entire ptrtbl block 1113 * can hold, then it'll be accounted for more than once, since 1114 * we won't have lastblk. 1115 */ 1116 for (i = 0; i < len; i++) { 1117 zap_leaf_t *l; 1118 1119 if (tbl[i] == lastblk) 1120 continue; 1121 lastblk = tbl[i]; 1122 1123 err = zap_get_leaf_byblk(zap, tbl[i], NULL, RW_READER, &l); 1124 if (err == 0) { 1125 zap_leaf_stats(zap, l, zs); 1126 zap_put_leaf(l); 1127 } 1128 } 1129 } 1130 1131 int 1132 fzap_cursor_move_to_key(zap_cursor_t *zc, zap_name_t *zn) 1133 { 1134 int err; 1135 zap_leaf_t *l; 1136 zap_entry_handle_t zeh; 1137 1138 if (zn->zn_key_orig_numints * zn->zn_key_intlen > ZAP_MAXNAMELEN) 1139 return (ENAMETOOLONG); 1140 1141 err = zap_deref_leaf(zc->zc_zap, zn->zn_hash, NULL, RW_READER, &l); 1142 if (err != 0) 1143 return (err); 1144 1145 err = zap_leaf_lookup(l, zn, &zeh); 1146 if (err != 0) 1147 return (err); 1148 1149 zc->zc_leaf = l; 1150 zc->zc_hash = zeh.zeh_hash; 1151 zc->zc_cd = zeh.zeh_cd; 1152 1153 return (err); 1154 } 1155 1156 void 1157 fzap_get_stats(zap_t *zap, zap_stats_t *zs) 1158 { 1159 int bs = FZAP_BLOCK_SHIFT(zap); 1160 zs->zs_blocksize = 1ULL << bs; 1161 1162 /* 1163 * Set zap_phys_t fields 1164 */ 1165 zs->zs_num_leafs = zap->zap_f.zap_phys->zap_num_leafs; 1166 zs->zs_num_entries = zap->zap_f.zap_phys->zap_num_entries; 1167 zs->zs_num_blocks = zap->zap_f.zap_phys->zap_freeblk; 1168 zs->zs_block_type = zap->zap_f.zap_phys->zap_block_type; 1169 zs->zs_magic = zap->zap_f.zap_phys->zap_magic; 1170 zs->zs_salt = zap->zap_f.zap_phys->zap_salt; 1171 1172 /* 1173 * Set zap_ptrtbl fields 1174 */ 1175 zs->zs_ptrtbl_len = 1ULL << zap->zap_f.zap_phys->zap_ptrtbl.zt_shift; 1176 zs->zs_ptrtbl_nextblk = zap->zap_f.zap_phys->zap_ptrtbl.zt_nextblk; 1177 zs->zs_ptrtbl_blks_copied = 1178 zap->zap_f.zap_phys->zap_ptrtbl.zt_blks_copied; 1179 zs->zs_ptrtbl_zt_blk = zap->zap_f.zap_phys->zap_ptrtbl.zt_blk; 1180 zs->zs_ptrtbl_zt_numblks = zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks; 1181 zs->zs_ptrtbl_zt_shift = zap->zap_f.zap_phys->zap_ptrtbl.zt_shift; 1182 1183 if (zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks == 0) { 1184 /* the ptrtbl is entirely in the header block. */ 1185 zap_stats_ptrtbl(zap, &ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), 1186 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap), zs); 1187 } else { 1188 int b; 1189 1190 dmu_prefetch(zap->zap_objset, zap->zap_object, 1191 zap->zap_f.zap_phys->zap_ptrtbl.zt_blk << bs, 1192 zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks << bs); 1193 1194 for (b = 0; b < zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks; 1195 b++) { 1196 dmu_buf_t *db; 1197 int err; 1198 1199 err = dmu_buf_hold(zap->zap_objset, zap->zap_object, 1200 (zap->zap_f.zap_phys->zap_ptrtbl.zt_blk + b) << bs, 1201 FTAG, &db, DMU_READ_NO_PREFETCH); 1202 if (err == 0) { 1203 zap_stats_ptrtbl(zap, db->db_data, 1204 1<<(bs-3), zs); 1205 dmu_buf_rele(db, FTAG); 1206 } 1207 } 1208 } 1209 } 1210 1211 int 1212 fzap_count_write(zap_name_t *zn, int add, uint64_t *towrite, 1213 uint64_t *tooverwrite) 1214 { 1215 zap_t *zap = zn->zn_zap; 1216 zap_leaf_t *l; 1217 int err; 1218 1219 /* 1220 * Account for the header block of the fatzap. 1221 */ 1222 if (!add && dmu_buf_freeable(zap->zap_dbuf)) { 1223 *tooverwrite += zap->zap_dbuf->db_size; 1224 } else { 1225 *towrite += zap->zap_dbuf->db_size; 1226 } 1227 1228 /* 1229 * Account for the pointer table blocks. 1230 * If we are adding we need to account for the following cases : 1231 * - If the pointer table is embedded, this operation could force an 1232 * external pointer table. 1233 * - If this already has an external pointer table this operation 1234 * could extend the table. 1235 */ 1236 if (add) { 1237 if (zap->zap_f.zap_phys->zap_ptrtbl.zt_blk == 0) 1238 *towrite += zap->zap_dbuf->db_size; 1239 else 1240 *towrite += (zap->zap_dbuf->db_size * 3); 1241 } 1242 1243 /* 1244 * Now, check if the block containing leaf is freeable 1245 * and account accordingly. 1246 */ 1247 err = zap_deref_leaf(zap, zn->zn_hash, NULL, RW_READER, &l); 1248 if (err != 0) { 1249 return (err); 1250 } 1251 1252 if (!add && dmu_buf_freeable(l->l_dbuf)) { 1253 *tooverwrite += l->l_dbuf->db_size; 1254 } else { 1255 /* 1256 * If this an add operation, the leaf block could split. 1257 * Hence, we need to account for an additional leaf block. 1258 */ 1259 *towrite += (add ? 2 : 1) * l->l_dbuf->db_size; 1260 } 1261 1262 zap_put_leaf(l); 1263 return (0); 1264 } 1265